The best upscaling process coming?

Hello,

Have you seen this? It looks impressive.

https://cinnafilm.com/cinnafilm-and-nvi ... onversion/

While other upscaling tools rely on generic algorithms or generative AI that creates hallucination errors

I suppose they think about Topaz.

Just a pr about something which others been doing for a while now. A lot typical buzz words
Example looks artificial and "plastic" like about all other AI upscalers, including Topaz.

Their Tachyon converter is quite good though.

Besides this, is there anything better than Topaz nowadays?

Hello,

Have you seen this? It looks impressive.

https://cinnafilm.com/cinnafilm-and-nvi ... onversion/

A quick Google search reveals this:

The Cinnafilm Tachyon plug-in for Harmonic Carbon ProMedia has an entry-level price of $6,000 per GPU instance/node in North America. Additionally, there's a fee of $2 and change for a 30-second ad using the Tachyon engine.

That's a bit high.

Harmonic Carbon ProMedia is EOL. That's an old article from 2013.

Tachyon engine is present in Telestream enterprise products and some other cloud based services. They license it. It's quite good. Maybe better than Alchemist.

Besides this, is there anything better than Topaz nowadays?

No idea. What can be good with Cinnafilm is that they have texture engine, so you can add some "grain" to make content look more realistic (although their sample on the grab looks "plastic").

I've tried Tachyon in Pixelstrings. There're no colorspace input for P3-DCI nor P3 in general

I thought it was a limitation in Alchemist only. Same issue with Tachyon. However, Rec.2020 is supported. I wonder if I should convert P3-DCI to Rec.2020 sources, apply the framerate conversion, and then reverse the colorspace from Rec.2020 to P3-DCI. But I'm not 100% sure the colors would stay intouched.

What would you suggest, Andrew?

Always work in Rec.2020, but with P3 limit. Then for any app your colorspace is Rec.2020 (fact it's P3 limited is meaningless). This way you have best compatibility. P3 is a 'special' colrospace, so better to use Rec.2020 with P3 limit. Still for safety you probably need to limit final master to P3 again, as processing may introduce colors outside P3 (not sure).

I know, I know... but my customers bring me P3-DCI masters because that's how they graded with their projectors. And they won't accept to remake them in Rec.2020 with the P3 limitation inside despite my advices. That's why I'm looking for a way to do the framerate conversions (with no time adjusment) by keeping their colors perfectly.

Hmmm. It's probably enough to make sure source and output interpretations are the same- they can be whatever. If they are the same then processing engine should not change gamut at all (well 95% sure). You can test if output file goes outside P3. If not it will work.
Wait- if source is YUV based and processing is RGB (which it will be if GPU is involved) then you need to know gamut and correct setting is important. So may not work. Try
If source/output is RGB based then gamut should not matter I think.

Conversion for cinema? They support different fps, doesn't have to be only 24p.

I know, I know... but my customers bring me P3-DCI masters because that's how they graded with their projectors. And they won't accept to remake them in Rec.2020 with the P3 limitation inside despite my advices. That's why I'm looking for a way to do the framerate conversions (with no time adjusment) by keeping their colors perfectly.

You can also ask Cinnafilm themselves how to handle it
They are quite responsive if I remember well.

I know about cinema... But clients sometimes ask framerate conversions for all their masters, including P3-DCI. I know it's a non-sense but they don't listen advices. If Tachyon definitely doesn't support P3, I'd use the 4% usual method.

I wrote to PixelStrings. Waiting for their answer.

Unless it's a classical music, 4% speed difference is better option than fps conversions.

“Hallucination errors” sounds like something from my color grade at 3AM

The advantage of frame rate conversion without changing the duration is that it avoids having to also convert things like 24 audio tracks and subtitle streams.

I just ran new tests with Alchemist. Your theory 'If source/output is RGB based then gamut should not matter I think.' seems correct on a 12-bit ProRes 4:4:4:4 sample. By leaving 'auto' in Alchemist's settings, the output appears to preserve the correct primary colors. However, the applied metadata is, of course, badly Rec.709. With Metacolor.Editor, I can change the primaries to P3 but not the EOTF to SMPTE 428M (that option isn't available in the list). I’ll need to find another piece of software for that.

EDIT: Looks like ffmpeg does the trick to re-apply the right metadatas.

Use
https://mogurenko.com/2021/01/29/amcdx- ... er-v0-6-7/

Not only you can change ( add, remove etc) metadata on MOV level, but also in ProRes private headers, so you get ‘perfect’ master.

Thanks!